WORKING PAPER
The Effect of Civil Conflict on Domestic Violence:
The Case of Peru
Italo A. Gutierrez and Jose V. Gallegos
RAND Labor & Population
WR-1168
September 2016
This paper series made possible by the NIA funded RAND Center for the Study of Aging (P30AG012815) and the RAND Labor and Population
Unit.
RAND working papers are intended to share researchers’ latest findings and to solicit informal peer review.
They have been approved for circulation by RAND Labor and Population but have not been formally edited or
peer reviewed. Unless otherwise indicated, working papers can be quoted and cited without permission of the
author, provided the source is clearly referred to as a working paper. RAND’s publications do not necessarily
reflect the opinions of its research clients and sponsors. RAND® is a registered trademark.
For more information on this publication, visit www.rand.org/pubs/working_papers/WR1168.html
Published by the RAND Corporation, Santa Monica, Calif.
© Copyright 2016 RAND Corporation
R ® is a registered trademark
Limited Print and Electronic Distribution Rights
This document and trademark(s) contained herein are protected by law. This representation of RAND
intellectual property is provided for noncommercial use only. Unauthorized posting of this publication online is
prohibited. Permission is given to duplicate this document for personal use only, as long as it is unaltered and
complete. Permission is required from RAND to reproduce, or reuse in another form, any of its research
documents for commercial use. For information on reprint and linking permissions, please visit
www.rand.org/pubs/permissions.html.
The RAND Corporation is a research organization that develops solutions to public policy challenges to help
make communities throughout the world safer and more secure, healthier and more prosperous. RAND is
nonprofit, nonpartisan, and committed to the public interest.
RAND’s publications do not necessarily reflect the opinions of its research clients and sponsors.
Support RAND
Make a tax-deductible charitable contribution at
www.rand.org/giving/contribute
www.rand.org
THE EFFECT OF CIVIL CONFLICT ON
DOMESTIC VIOLENCE:
THE CASE OF PERU
Italo A. Gutierrez*
RAND Corporation
Jose V. Gallegos
Universidad de Piura
First Draft: August 2011; This Revision: September 2016
Abstract
We study the effect of women’s exposure to civil conflict violent events during childhood
and early teenage years on the probability that they will experience domestic violence in their
marriages as adults. In particular, we investigate the case of the internal conflict in Peru during
the 1980s and early 1990s, and its effect on the incidence of domestic violence between 2004
and 2012. We find that female exposure to conflict violence increases their later risk of being a
perpetrator and a victim of domestic violence. The average effects for women affected by the
conflict are small, although they mask important heterogeneities as some regions were affected
by the conflict more severely than others. The effects are substantial for women in the highest
categories of exposure. We also find evidence that a potential mechanism through which
exposure to the conflict affects domestic violence in the long-term is normalization of the use of
violence. Women more exposed to conflict violent events are more likely to justify the use of
violence against women and more likely to stay in a violent relationship.
Keywords: domestic violence, civil conflicts, social norms, Peru
JEL Classification Codes: J12; J16;
*
Corresponding author. Email: [email protected]; Address: 1776 Main Street, Santa Monica, CA, 90407, USA
1
1. Introduction
There is ample evidence that domestic violence is more likely to occur among persons
who were exposed to domestic violence as children. Many studies have provided support for the
link between growing up in a violent family environment and being a perpetrator of spousal
abuse (Barnett, Miller-Perrin and Perrin, 2005, Caesar, 1988, Stith et al., 2000), as well as being
a victim of spousal abuse (Cappell and Heiner, 1990, Mihalic and Elliott, 1997, Simons et al.,
1993, Stith et al., 2000). In contrast, the relationship between exposure to any type of community
violence and experience of subsequent domestic violence
has been less well-established
empirically (Buvinic, Morrison and Shifter, 1999). This study is one of the first to examine this
relationship. In particular, we study the effect of women’s exposure to civil conflict events
during their childhood and early teenage years on their probability of being perpetrators and
victim of domestic violence in their adult relationships. We do this analysis for the case of Peru,
by linking pooled cross-sections of the Peruvian Demographic and Health Survey (DHS) across
the years 2004 through 2012, which provides information on domestic violence, with a detailed
registry of conflict-related events between 1980 and 2000.
Civil conflicts are known to have long term effects. In the Peruvian case, previous
research has found effects on schooling, physical height and earnings capacity. Leon (2012)
found that the average person exposed to the political violence in Peru before starting school
accumulated about 0.21 less years of education by adulthood. This effect is more important for
women than for men and for Spanish speakers than for native speakers. Sanchez (2010) found
evidence that the conflict had an impact on child nutrition. He found that in regions that were
affected by a high level of violence, a one percent increase in conflict intensity reduced average
1
height-for-age by about 3% of the standard deviation. Grimard and Laszlo (2014) found that the
effects of the conflict on height persist in the long term, at least in the case of women. Galdo
(2010) found that a one standard deviation increase in civil war exposure during the first 36
months of life leads to a 4% fall in adult monthly earnings.
Our study contributes to this literature by examining another possible long-term outcome
of the civil conflict in Peru. Our empirical investigation is based upon the hypothesis that early
exposure to violence in the community can increase the likelihood of experiencing violence
within households. Although there is no prior direct evidence of this link in the psychology
literature, there are many studies that suggest that violence is a learned behavior (or a social
norm). These studies would support our prediction that people that were exposed to conflict
violence in their childhood may be more likely to resort to violence when dealing with their own
partners. For example, Fowler et al. (2009) states that “Social cognition theories propose that
exposure to community violence normalizes the use of aggressive behavior. As a result, youths
learn that violence is an effective method of problem solving and, therefore, are more likely to
engage in violent acts themselves” (p. 228). Guerra, Huesmann and Spindler (2003) also find
that exposure to community violence among urban elementary school children in Chicago
increases normative beliefs approving aggression and aggressive fantasies. Schwab-Stone et al.
(1995) find that, among 6th, 8th and 10th graders, exposure to violence is associated with greater
willingness to use physical aggression, diminished perception of risk (of risky activities), lower
personal expectations, alcohol use and diminished academic achievement. Scarpa, Haden and
Hurley (2006) explore the consequences of lifetime community violence exposure and find that
the incidence of depression, post-traumatic stress symptoms, aggressive behavior, and
2
interpersonal problems is higher among individuals who were exposed to community violence
during their early adulthood.
In addition, there is evidence that not only being a victim, but witnessing or hearing about
violent events also impacts child development (Fowler et al., 2009, Scarpa et al., 2006). Living
in constant fear of violence can result in severe psychological damage and affect child
development. Irritability, interpersonal conflict and anger are usually mentioned as normal
reactions to trauma. According to Williams (2007) , “the negative developmental effects appear
more likely if children experience repeated or repetitive 'process' trauma or live in unpredictable
climates of fear” (p. 274). This evidence is important for our research because even though we
cannot identify in our data direct victims of the armed conflict in Peru, we can measure the level
of violence in their area of residence.
Although this is not the only paper that studies the effects of civil conflicts on domestic
violence, it is the first that focuses on the effects of exposure in early childhood and teenage
years.1 Grimard and Laszlo (2014) also focus on the Peruvian case, but they focus on in-utero
exposure to the conflict (until 12 months after birth) and find no effects. Noe and Rieckmann
(2013) analyze the case of Colombia, finding that higher conflict intensity increases the
likelihood of women becoming a victim of domestic violence. However, they investigate the
effect of contemporary violence on domestic violence, while our study evaluates the long-term
impact of early exposure on the incidence of domestic violence in adult relationships. La-Mattina
(2015) investigates the case of the Rwandan genocide. She finds that women who married after
the 1994 genocide experience significantly increased domestic violence than women who
1
It is worth noting that the papers cited here were developed contemporaneously with our study. Some of them
(La-Mattina, 2015, Noe and Rieckmann, 2013) cite an early version of this study.
3
married before. La-Mattina (2015) also explores long term effects, as she uses the 2005 Rwanda
Demographic and Health Survey (DHS). However, unlike the Peruvian case, the Rwanda
conflict was a high intensity, short duration conflict that substantially changed the structure of
the population. As a result, La-Mattina (2015) finds that part of the effect of the genocide on
domestic violence is driven by changes in the marriage market sex ratios induced by the conflict.
The Peruvian case provides a different context, since it was a longer conflict (for over a decade)
and its intensity (measured by total related deaths and disappearances) were not as large as in the
Rwandan genocide. Hence, we do no find evidence that in the Peruvian case the long-term
effects of the conflict on domestic violence operates through the marriage market. Also, as noted
earlier, an important difference is that we investigate the effects of the conflict not for women of
marriageable age, but for women exposed to the conflict as children and early teenagers.
Our results indicate that women’s exposure to civil conflict violent events during their
childhood and early teenage years is associated with increases in the probability of being a
perpetrator of violence towards their partners and in the probability of being a victim of violence
from their partners. The sizes of the effects are small for the average woman affected by the
conflict. However, we find substantial effects for women with high levels of exposure to the
conflict. For example, a level of exposure corresponding to the highest decile increases the
probability of a woman being victim of any kind of violence (psychological, physical or sexual)
in the last 12 months by 11 percentage points. This is a substantial effect, given that the sample
average is 22%.
The rest of the paper is organized as follows. Section 2 of this paper presents a brief
description on the evolution of the civil conflict in Peru during the 1980s and 1990s. Section 3
describes the data sources. Section 4 discusses the empirical approaches and presents the
4
estimation results. Section 5 discusses potential mechanisms through which exposure to conflict
violence at an early age can affect domestic violence later in life. Section 6 presents a number of
robustness checks. Section 7 discusses the data limitations and their implication for our estimated
results. Finally, Section 8 presents the conclusions of this study.
2. The Civil Conflict in Peru
The aim of this section is to provide a brief recount of the evolution of the conflict in
Peru. It started in 1980 when the Peruvian Communist Party-Shining Path (PCP-SL) declared
war against the Peruvian State. The symbolic action that started the conflict was the burning of
the election ballots in the district of Chuschi, in the region of Ayacucho (south-central highlands
of Peru), in May 1980 (TRC 2004). Initially, the PCP-SL carried out isolated attacks against
public and private property and different propaganda actions in favor of the armed conflict. The
government perceived them as marginal events, concentrated in the region of Ayacucho and
perpetrated by a small group of left wing insurgents. However, the level of violence of the
attacks systematically escalated between 1980 and 1982 (see Figure 1). The government
answered by increasing the severity of its response and by 1983 the Peruvian army joined the
war against the PCP-SL. The level of violence and unrest spiked in 1983-1984 and then again in
1989-1990 (Figure 1). In 1984, the Revolutionary Movement Tupac Amaru (MRTA), a selfproclaimed left-wing group, also joined the war against the State.
By 1986, the conflict had spread to other parts of the country including the regions of
Puno, Junin, the Huallaga Valley and Lima, the capital of Peru. By the early 1990s the violence
peaked in the capital, especially in the form of murders and terrorists attacks. In 1991, more than
half of the population in Peru lived under a curfew system with restricted civil rights.
5
The PCP-SL not only targeted popular leaders, local authorities and land holders, but also
attacked the civil population. Part of the strategy of PCP-SL was to recruit individuals using
ideological propaganda and a strategy of intimidation. Thus, whenever residents of a village
declared themselves against the revolution or did not want to provide food or supplies, they were
brutally punished (Leon, 2012). As a result, the vast majority of the victims of human rights
violations during the conflict were civilians. Farmers accounted for almost half of the victims
(48%), while local traders, independent workers and housewives accounted for an additional
20% (Leon, 2012).
In September 1992 the leader of the PCP-SL, Abimael Guzman, was captured. In the
years that followed the level of violence decreased rapidly. A peace agreement was proposed by
Guzman in October 1993. Although an agreement was not reached, it was a key factor in the
Government propaganda strategy and many militants of the PCP-SL decided to abandon the
armed conflict against the State. Additional arrests of important leaders from the PCP-SL and the
MRTA further weakened these movements. By the second half of the 90s the conflict had
largely subsided with only a few significant isolated violent events taking place. In total,
according to the Truth and Reconciliation Commission (TRC) nearly 70,000 people died in Peru
between 1980 and 2000 due to the civil conflict.
3. Data Description
We use two main datasets in this study. Information on domestic violence comes from
the Peruvian Demographic and Health Survey (DHS), across years 2004 through 2012 (pooled
yearly cross-sections). The DHS is a nationally-representative sample of women between 15 and
49 years of age. The DHS includes an extensive set of questions to determine if the respondent is
6
(or has been) a victim of domestic violence. Also, it is perhaps one of the few surveys that collect
this information using a representative population sample rather than using, for example, data
from shelters for abused women. In the Peruvian DHS, one woman per household is randomly
selected to answer the domestic violence questionnaire, which is the last section of the interview.
The available information in the DHS allows us to distinguish between three types of violence
against women: psychological, physical and sexual violence. Psychological violence
encompasses verbal abuse, threats of physical violence or threats about leaving or withdrawing
economic support. Physical violence includes different forms of battering and attacks (or threats)
with a knife, gun or other weapons. Sexual violence is defined as having sexual relations against
the partner’s will. The DHS also contains questions regarding physical violence from the woman
towards her husband or partner. The Appendix shows how the questions available in the DHS
regarding domestic violence are used in this study.
A limitation of the DHS for this study is the lack of complete information about women's
migratory history. In order to construct a reliable measure of women exposure to conflict
violence during childhood and early teenage years, we select for our analysis women who report
to have always lived in the same town, i.e. non-migrant women, which represents about half of
the sample. This selection can introduce bias in our results since non-migrant women may be
different from women who migrated. For example, women that migrated may have been affected
more severely by the conflict. Thus, our analysis would only include women that were not
displaced by the conflict. We return to this point in Section 7.
The second dataset are detailed records on conflict-related events that occurred in Peru
between 1980 and 2000. These records were collected by the Peruvian Truth and Reconciliation
Commission (TRC, 2004) and provide information on the location, year, type, victims, and
7
perpetrator group associated with each event. This allows us to construct variables measuring
exposure to violent events related to the civil conflict. We define a violent event as an
assassination, a kidnapping, a sexual assault, or an attack on a public or private facility or
institution that occurred in a particular province of Peru. We aggregate events at the province
level rather than at a lower level (i.e. district) because the DHS is not representative at low levels
of geographic disaggregation. Peru is comprised of 24 regions, 194 provinces and 1,818 districts.
The DHS is designed to be representative only at the regional level. Moreover, since only one
woman per household in the DHS is selected for the domestic violence questionnaire, the
number of available observations is smaller than in the full sample. However, since we are
pooling many years of data, we are able to perform inference at lower level of geographic
aggregation. In our analysis, we aggregate conflict-related violent events at the province level as
compromise between measuring localized exposure to conflict-related violence and the statistical
representativeness of the data.
Based upon the timing of the conflict-related events, the woman’s year and province of
birth, we constructed her early exposure to conflict violence, following a methodology similar to
Leon (2012). Our measure records the total number of conflict violent events that occurred in the
woman’s province of residence until the age of 16. We stopped at age 16 because we want to
restrict our measure of exposure to conflict violent events to the years prior to the marriageable
age, to be certain that the exposure has occurred prior to marriage.2
2
In Peru, even though the majority is reached at age 18, individuals can get married at the age of 16 with parental
consent. In fact, we observe in the data that a non-trivial fraction of women are married as early as age 16.
8
4. Empirical strategy and estimation results
The empirical strategy relies on the variation in the timing and frequency of conflict
violent events over time and across provinces. The civil conflict reached different geographic
areas in different years and with different intensities over time as was described in Section 2.
Figure 1 provides two examples. It shows the incidence of deaths and reported missing people in
the region of Ayacucho, where the conflict started and one of the most affected regions, and the
region Lima, the capital of Peru. Figure 1 shows that conflict-related violence peaked in
Ayacucho by 1984, to slowly decline after that (as it spread to other regions in the country). In
comparison, in Lima the violence peaked relatively late, in the early 90s, and declined towards
the end of the decade.
Figure 1: Number of Deaths and Missing People
Ayacucho
Lima
3000
120
2500
100
2000
80
1500
60
1000
40
500
20
0
0
1980
1984
1988
1992
Ayacucho
1996
2000
Lima
Source: TRC (2004)
We use this geographic and temporal variation in the occurrence of conflict violent
events to identify their effects on the probability of domestic violence in later-life relationships.
Our main specification is the linear probability model shown in equation (1):
‫ݕ‬௜ǡ௝ǡ௧ǡ௪ ൌ ߙ ൅ ߚ‫ܧܸܥ‬௜ǡ௝ǡ௧ ൅ ߛܺ௜ ൅ ߜ௝ ൅ ߶௧ ൅ ߣ௪ ൅ ߳௜
9
(1)
The subscript i indexes individuals, the subscript j indexes provinces, the subscript t
indexes year of birth, and the subscript w indexes wave or year of interview. The variable
‫ܧܸܥ‬௜ǡ௝ǡ௧ measures the total exposure to conflict violent events up to the age of 16, our main
variable of interest. The term ܺ௜ is a vector that controls for other factors that can affect the
occurrence of domestic violence. In particular, vector ܺ௜ controls for: i) violence observed at
home as a child (whether women were battered by their fathers or by their mothers or whether
their fathers used to batter their mothers); ii) education attainment (the woman’s education level,
her husband or partner’s education level and whether the education level of the woman is lower
than her husband’s as a measure of being in a disadvantaged position in the household); iii) labor
force status and labor income (whether the woman works for pay and whether woman’s income
is lower than her husband’s income as a measure of being in a disadvantaged position in the
household); iv) women’s body mass index (BMI); v) age at which the woman got married or
started the union; vi) the woman's ethnicity; vii) the household’s wealth index; and, viii) the
number of children five years old or younger. We also control for province effects (term ߜ௝ ), for
cohort or birth year effects (term ߶௧ ) and for year of interview effects (term ߣ௪ ).
We investigate the effect of early exposure to conflict violent events on several measures
of domestic violence, represented by the dependent variable ‫ݕ‬௜Ǥ௝ǡ௧ǡ௪ in equation (1). In particular,
we look at the effects on the following variables: i) physical violence from the woman to her
partner; ii) physical violence from partner to the woman; iii) sexual violence from the partner to
the woman; iv) psychological violence from the partner to the woman. The Appendix shows how
each of these variables is measured in the DHS. The impacts of early exposure to conflict violent
events are estimated for whether these measures of domestic violence have ever occurred and for
10
whether they have occurred within the last 12-months prior to the interview. Thus, we estimate
not only if exposure to conflict violence (in the pre-marriageable years) has ever affected the
incidence of domestic violence within the marriage, but also whether exposure to conflict
violence continues to affect the incidence of domestic violence in the long term, i.e. more than
20 years after the start of the conflict.
Table 1 shows the sample averages of the outcome variables and of exposure to conflict
violent events. Table 1 shows that about 10% of women report to have physically abused their
partners at some point. In comparison, 39% of women report to have been physically abused by
their partners at some point. Also, about 32% and 9% of women report to have been abused
psychologically or sexually, respectively, by their partners. In total, almost half of the women in
the sample (47%) report to have been abused (physically, sexually, or psychologically) by their
partners at some point. Table 1 also shows that the incidence of domestic violence in the last 12
months prior to the interview is smaller (as expected) but still large. Almost one quarter (22%) of
the women in the sample report to have been victims of some type of violence (physical, sexual,
or psychological) within the last year. The prevalence of domestic violence in Peru is
incontestably large. In fact, Peru is one of the countries with the highest rates of violence against
women in the Latin American and Caribbean region, only below Bolivia and similar to Colombia
(ECLAC, 2015). Table 1 also shows that women in the sample were exposed on average to 28
violent events related to the civil conflict when they were between 0 and 16 years old. It also
shows that this variable has very extreme values, indicating that some areas were affected more
severely by the conflict than others.
Estimation of equation (1) is limited to provinces with at least one conflict-related violent
event, since the identification strategy relies on variation in the amount of exposure among
11
women born in the same province at different times. We also limit the sample to women born
between 1965 and 1990, so that all women in our sample were 15 years old or younger when the
internal conflict started, and are of marriageable age by the start of the estimation sample (i.e.
2004). Since we are studying domestic violence or intimate partner violence, we also limit our
sample to women who are married (or cohabitate with a partner) or were married (cohabitated) in
the past. Standard errors in equation (1) are clustered at the province level. Table 2 shows the
estimation results. We report the coefficient associated with early exposure to conflict violent
events up to the age of 16 (CVE). For comparison purposes, we also report the coefficients
associated with measures of violence observed at home as a child. These variables are strong
predictors of domestic violence in later life, both in our data and in the literature.
12
Table 1: Descriptive Statistics of Outcomes and Exposure to Conflict-Violence
Variable
Mean
Std.
Dev
Min
Max
Domestic violence against women - Ever (0=No; 1=Yes)
Physical violence from woman to partner
0.10
0.29
0.00
1.00
Physical violence from partner to woman
0.39
0.49
0.00
1.00
Sexual violence from partner to woman
0.09
0.28
0.00
1.00
Psychological violence from partner to woman
0.32
0.47
0.00
1.00
Any kind of violence from partner to woman
0.47
0.50
0.00
1.00
Domestic violence against women - Last 12 months (0=No; 1=Yes)
Physical violence from woman to partner
0.04
0.19
0.00
1.00
Physical violence from partner to woman
0.14
0.35
0.00
1.00
Sexual violence from partner to woman
0.04
0.18
0.00
1.00
Psychological violence from partner to woman
0.17
0.37
0.00
1.00
Any kind of violence from partner to woman
0.22
0.41
0.00
1.00
0.28
0.67
0.00
4.52
0.12
0.42
0.00
4.10
0.16
0.41
0.00
3.90
Exposure to Civil Conflict Violence
Total CVE between 0 and 16 years old (in
hundreds)
Total CVE between 0 and 8 years old (in
hundreds)
Total CVE between 9 and 16 years old (in
hundreds)
Placebo: Total CVE between minus 8 and 0
years old (in hundreds)
0.00
3.57
0.02
0.20
Source: DHS (2004-2012)
Notes: Includes only non-migrant women from provinces affected by the internal conflict, born
in 1965-1990 (see text for more details).
13
Table 2: Regression Results - Effect of Early Exposure to Conflict Violent Events on Domestic Violence Outcomes
Physical violence from
woman to partner
(0=No; 1=Yes)
Physical violence from
partner to woman
(0=No; 1=Yes)
Sexual violence from
partner to woman
(0=No; 1=Yes)
Psychological violence
from partner to woman
(0=No; 1=Yes)
Any type of violence
from partner to woman
(0=No; 1=Yes)
Ever
Last 12
months
Ever
Last 12
months
Ever
Last 12
months
Ever
Last 12
months
Ever
0.010*
0.010**
0.018*
0.012**
0.003
0.000
0.019*
0.019*
0.027***
0.022*
(0.006)
(0.004)
(0.010)
(0.006)
(0.009)
(0.004)
(0.010)
(0.011)
(0.009)
(0.012)
0.037***
(0.013)
0.076***
0.051***
0.021*
0.009
0.075***
0.057***
0.094***
0.072***
(0.017)
(0.015)
(0.012)
(0.008)
(0.017)
(0.016)
(0.019)
(0.020)
0.031***
0.072***
0.042***
0.046***
0.019**
0.075***
0.048***
0.089***
0.060***
(0.011)
(0.009)
(0.013)
(0.014)
(0.012)
(0.009)
(0.017)
(0.016)
(0.014)
(0.018)
0.041***
0.014***
0.147***
0.062***
0.036***
0.016***
0.098***
0.059***
0.156***
0.083***
(0.005)
(0.003)
(0.006)
(0.006)
(0.005)
(0.002)
(0.007)
(0.006)
(0.006)
(0.006)
17,270
17,261
17,270
17,269
17,269
17,269
17,270
17,267
17,270
17,270
Last 12
months
Early Exposure to Civil Conflict Violence
Total CVE between 0 and
16 years old (in hundreds)
Exposure to Violence at Home as a child
Woman battered by mother
0.057***
when child
(0.016)
Woman battered by father
0.063***
when child
Father used to batter mother
# Observations
Notes: Standard errors are clustered at the province level. Includes only non-migrant women from provinces affected by the internal conflict, born in 1965-1990. Each regression
includes controls for province effects, year of birth effects, and interview year effects. We also control for the following characteristics: i) violence observed at home as a child
(whether women were battered by their fathers or by their mothers or whether their fathers used to batter their mothers); ii) education attainment (women’s education level, their
husband or partners’ education level, and whether the education level of the woman is lower than her husband’s) iii) labor force status and labor income (whether the women
works for pay and whether woman’s income is lower than her husband’s income); iv) women’s BMI; v) age at which the women got married or started the union; vi) women's
ethnicity; vii) household’s wealth index; and, viii) the number of children five years old or younger in the household. *** denotes statistical significance at 1% level; ** denotes
statistical significance at 5% level; * denotes statistical significance at 10% level.
14
The results in Table 2 indicate that early exposure to conflict violent events increases the
probability of physical violence from the woman to her partner and from the partner to the
woman. It also increases the probability of psychological violence against the woman, although
the effects are statistically significant at the 10% level. We find no statistically significant effect
of early exposure to conflict violence on sexual violence. Interestingly, Table 2 also shows that
the effects of exposure do not seem to lessen much over time. The effects are similar in size
regardless whether we measure the occurrence of domestic violence at any point in time (ever) or
in the last 12 months (the small differences in size are not statistically significant).
We can also assess the importance of the civil conflict as a determinant of the incidence
of domestic violence, using the estimation results in Table 2. The average woman in the sample
was exposed to 28 violent events prior to the age of 16. Using the estimates from the last two
columns of Table 2, back-of-the-envelope calculations indicate that the conflict increased on
average by 0.8 percentage points (i.e. 0.027 x 0.28) the probability of ever occurring any type of
violence against a woman from her partner, and by 0.6 percentage points (i.e. 0.022 x 0.28) the
probability of its occurrence in the last 12 months. These average effects are small. In
comparison, following a similar exercise, we calculate that the average effect of exposure to
violence at home during childhood, in particular of seeing their fathers battering their mothers
(which has a prevalence of 47% in the sample), is 7.3 percentage points on the probability of any
type of violence ever against the woman from her partner, and 3.9 percentage points on the
probability of occurrence in the last 12 months.
The small average effect of the conflict on domestic violence masks the large
heterogeneity in the total exposure to conflict-related violence. To investigate how the conflict
affected the incidence of domestic violence for those who were more exposed to conflict violent
15
events, we recode the early exposure variable ‫ ܧܸܥ‬into several binary variables ‫ܦܧܸܥ‬௞ , which
equal one if the level of exposure belongs to the ݇ ௧௛ decile category in the distribution of ‫ܧܸܥ‬,
and equal zero otherwise. We then estimated the linear probability model in equation (2) below,
where the rest of the terms are defined as in equation (1).
௞
‫ݕ‬௜ǡ௝ǡ௧ ൌ ߙ ൅ σଵ଴
௞ୀଵ ߚ௞ ‫ܦܧܸܥ‬௜ǡ௝ǡ௧ ൅ ߛܺ௜ ൅ ߜ௝ ൅ ߶௧ ൅ ߣ௪ ൅ ߳௜
(2)
The estimation results are shown in Table 3. It also presents the average exposure to
conflict violent events in each decile category. The exposure in the first two decile categories is
zero events, so they form the omitted group. As expected, most of the statistically significant
effects are positive. The only exception is the effect for the 4th decile category in the regression
of ever occurring physical violence against the partner. More importantly, the effects of exposure
increase substantially at higher decile categories. For instance, we find that a level of exposure
corresponding to the highest decile category (and average of 186 violent events) increased the
probability of physical violence against the woman from her partner in the last 12 months by 7
percentage points. Given that the sample mean is 14% (Table 1), this is a substantial effect. We
observe similar large effects in the probability of psychological violence (ever and in the last 12
months) and in the probability of any type of violence (ever and in the last 12 months).
Moreover, we also find important effects for the probability of sexual violence. Women exposed
to the highest decile of violence are 5.1 percentage points more likely to have ever been a victim
of sexual violence from their partners. Given that the sample mean is 9% (Table 1), this is also a
substantial effect.
16
Table 3: Effect of Early Exposure to Conflict Violent Events on Domestic Violence Outcomes, By Decile of Exposure
Decile of
CVE
Exposure
3rd decile
Average
exposure
(# events
in
hundreds)
0.010
4th decile
0.020
5th decile
0.030
6th decile
0.048
7th decile
0.114
8th decile
0.216
9th decile
0.592
Highest
decile
# of
Observations
1.86
Physical violence from
woman to partner
(0=No; 1=Yes)
Ever
Last 12
months
Physical violence from
partner to woman
(0=No; 1=Yes)
Ever
Last 12
months
Sexual violence from
partner to woman
(0=No; 1=Yes)
Ever
Last 12
months
Psychological violence
from partner to woman
(0=No; 1=Yes)
Ever
Last 12
months
Any type of violence
from partner to woman
(0=No; 1=Yes)
Ever
Last 12
months
0.003
0.009
0.027
0.022*
0.024**
0.004
0.033**
0.011
0.028*
0.031**
(0.010)
(0.007)
(0.017)
(0.012)
(0.009)
(0.007)
(0.015)
(0.014)
(0.016)
(0.014)
-0.027***
-0.004
0.045***
0.012
0.031***
0.005
0.065***
0.027*
0.057***
0.029
(0.009)
(0.006)
(0.016)
(0.015)
(0.011)
(0.006)
(0.015)
(0.016)
(0.015)
(0.020)
-0.008
-0.003
0.021
0.008
0.029***
-0.003
0.058**
0.040**
0.032
0.030
(0.011)
(0.007)
(0.023)
(0.018)
(0.011)
(0.007)
(0.026)
(0.019)
(0.026)
(0.021)
-0.020
-0.002
0.030
0.032*
0.019
-0.006
0.040
0.019
0.041
0.033
(0.013)
(0.007)
(0.027)
(0.018)
(0.013)
(0.009)
(0.029)
(0.020)
(0.027)
(0.020)
-0.016
-0.008
0.026
0.026*
0.025**
0.004
0.016
0.028*
0.018
0.037**
(0.014)
(0.006)
(0.020)
(0.015)
(0.012)
(0.007)
(0.019)
(0.015)
(0.019)
(0.018)
-0.012
-0.002
0.043*
0.039**
0.015
0.008
0.050*
0.030
0.053**
0.058**
(0.016)
(0.010)
(0.024)
(0.018)
(0.014)
(0.008)
(0.026)
(0.021)
(0.022)
(0.022)
-0.005
0.002
0.019
0.039
0.027
-0.002
0.055*
0.056**
0.040
0.070**
(0.016)
(0.013)
(0.039)
(0.026)
(0.017)
(0.010)
(0.031)
(0.027)
(0.029)
(0.029)
0.012
0.010
0.058
0.070***
0.051**
0.012
0.087***
0.081***
0.082***
0.111***
(0.018)
(0.011)
(0.036)
(0.023)
(0.020)
(0.012)
(0.032)
(0.029)
(0.029)
(0.029)
17,270
17,261
17,270
17,269
17,269
17,269
17,270
17,267
17,270
17,270
Notes: Standard errors are clustered at the province level. Includes only non-migrant women from provinces affected by the internal conflict, born in 1965-1990. Each
regression includes controls for province effects, year of birth effects, and interview year effects. We also control for the following characteristics: i) violence observed at home
as a child (whether women were battered by their fathers or by their mothers or whether their fathers used to batter their mothers); ii) education attainment (women’s education
level, their husband or partners’ education level, and whether the education level of the woman is lower than her husband’s) iii) labor force status and labor income (whether the
women works for pay and whether woman’s income is lower than her husband’s income); iv) women’s BMI; v) age at which the women got married or started the union; vi)
women's ethnicity; vii) household’s wealth index; and, viii) the number of children five years old or younger in the household. *** denotes statistical significance at 1% level;
** denotes statistical significance at 5% level; * denotes statistical significance at 10% level.
17
We also investigate whether the timing of exposure to conflict violence during a
woman’s childhood and teenage years has differential effects on the probability of domestic
violence in their later life. To study this issue, we use again the empirical specification in
equation (1) but we divide the total exposure to conflict violent events in two segments: exposure
between 0 and 8 years of age (CVE8) and exposure between 9 and 16 years of age (CVE16), as
shown in equation (3) below. Table 1 shows that the average exposure to 28 conflict violent
events between the ages of 0 and 16 is divided in 12 violent events between the ages 0 and 8, and
16 violent events between the ages of 9 and 16.4 Besides the exposure to conflict violent events
between the ages of 0-8 and 9-16, we also include two additional regressors in equation (3).
First, an interaction term between the demeaned values of ‫ܧܸܥ‬ͺ and ‫ͳܧܸܥ‬͸, to allow for the
effect of exposure to conflict violence between ages 9 and 16 on domestic violence to be
moderated by the effect of exposure at earlier ages (0 to 8), and vice versa. Second, an additional
regressor, ‫ݎ݋݅ݎ̴ܲܧܸܥ‬ͺ, which measures the total number of conflict violent events in the
province of birth during the 8 years prior to the birth of the woman. This term is included to act
as a placebo to test the validity of the causal interpretation of our estimates. If we are identifying
the effect of exposure to conflict violence, then conflict violent events that occurred prior to birth
of the woman should not affect the incidence of domestic violence later in life. If, on the
contrary, we find an effect of this placebo variable of the probability of domestic violence, it
would cast doubt on the interpretation of our findings as the effects of exposure to conflict
violence.
4
The average increase in exposure to violent event at later ages seems to be at odds with the fact that the conflict
died down after 1992. However, this is just an artifact of the age composition in the sample. The average birth year
in the sample is 1976, which means that the average woman was in her teenage years when the conflict violence
peaked by the end of the 80s
18
തതതതതതതതതതത
‫ݕ‬௜ǡ௝ǡ௧ ൌ ߙ ൅ ߚଵ ‫ܧܸܥ‬ͺ௜ǡ௝ǡ௧ ൅ ߚଶ ‫ͳܧܸܥ‬͸௜ǡ௝ǡ௧ ൅ ߚଷ ൣ൫‫ܧܸܥ‬ͺ௜ǡ௝ǡ௧ െ ‫ܧܸܥ‬ͺ
పǡఫǡ௧ ൯ ൈ ൫‫ͳܧܸܥ‬͸௜ǡ௝ǡ௧ െ
തതതതതതതതതതതതത
‫ͳܧܸܥ‬͸పǡఫǡ௧ ൯൧ ൅ ߚସ ‫ݎ݋݅ݎ̴ܲܧܸܥ‬ͺ௜ǡ௝ǡ௧ ൅ ߛܺ௜ ൅ ߜ௝ ൅ ߶௧ ൅ ߣ௪ ൅ ߳௜
(3)
Table 4 shows the estimation results of equation (3). Although the majority of the
coefficients for ‫ܧܸܥ‬ͺ are larger than the coefficients for ‫ͳܧܸܥ‬͸, the differences are not large
enough to be statistically significant. Moreover, the estimates for the summary measures of any
type of violence (ever and in the last 12 months) are very similar for ‫ܧܸܥ‬ͺ and for ‫ͳܧܸܥ‬͸.
Thus, there is no evidence that exposure to conflict violence at an earlier (or later) age has a
differential impact on the incidence of domestic violence in later life. Table 4 also shows that
the estimated coefficients for ‫ݎ݋݅ݎ̴ܲܧܸܥ‬ͺ are much smaller (in many cases by an order of
magnitude) than the coefficients for ‫ܧܸܥ‬ͺ and ‫ͳܧܸܥ‬͸. Also, none of the coefficients for
‫ݎ݋݅ݎ̴ܲܧܸܥ‬ͺ are statistically significant. As explained above, this result lends more credence to
the interpretation of our findings as the causal effect of exposure to conflict-related violence.
19
Table 4: Effect of Early Exposure to Conflict Violent Events on Domestic Violence Outcomes, By Age of Exposure
Physical violence from
woman to partner
(0=No; 1=Yes)
Total CVE between 0 and 8
years old
Total CVE between 9 and
16 years old
Interaction term
Physical violence from
partner to woman
(0=No; 1=Yes)
Sexual violence from
partner to woman
(0=No; 1=Yes)
Psychological violence
from partner to woman
(0=No; 1=Yes)
Any type of violence
from partner to woman
(0=No; 1=Yes)
Ever
Last 12
months
Ever
Last 12
months
Ever
Last 12
months
Ever
Last 12
months
Ever
Last 12
months
0.016*
(0.008)
0.016**
(0.007)
0.003
(0.013)
0.019**
(0.008)
-0.002
(0.013)
0.007
(0.006)
0.022
(0.017)
0.040***
(0.015)
0.021
(0.018)
0.038**
(0.015)
0.020**
0.010*
0.010
0.013
0.008
0.001
0.015
0.026**
0.020
0.034***
(0.008)
(0.006)
(0.017)
(0.012)
(0.012)
(0.005)
(0.013)
(0.013)
(0.012)
(0.013)
-0.009
-0.009
0.017**
-0.012
0.002
-0.004
-0.009
-0.022**
-0.000
-0.025**
(0.006)
(0.006)
(0.008)
(0.010)
(0.008)
(0.003)
(0.010)
(0.011)
(0.014)
(0.011)
Placebo: Total CVE
between minus 8 and 0 years
old
0.007
-0.011
-0.006
-0.014
0.003
0.008
-0.023
0.006
-0.028
-0.008
(0.010)
(0.006)
(0.029)
(0.016)
(0.012)
(0.009)
(0.021)
(0.014)
(0.028)
(0.018)
# of Observations
17,270
17,261
17,270
17,269
17,269
17,269
17,270
17,267
17,270
17,270
Notes: Standard errors are clustered at the province level. Includes only non-migrant women from provinces affected by the internal conflict, born in 1965-1990. Each
regression includes controls for province effects, year of birth effects, and interview year effects. We also control for the following characteristics: i) violence observed at
home as a child (whether women were battered by their fathers or by their mothers or whether their fathers used to batter their mothers); ii) education attainment (women’s
education level, their husband or partners’ education level, and whether the education level of the woman is lower than her husband’s) iii) labor force status and labor income
(whether the women works for pay and whether woman’s income is lower than her husband’s income); iv) women’s BMI; v) age at which the women got married or started
the union; vi) women's ethnicity; vii) household’s wealth index; and, viii) the number of children five years old or younger in the household. *** denotes statistical
significance at 1% level; ** denotes statistical significance at 5% level; * denotes statistical significance at 10% level.
20
Besides exposure to conflict violence and to violence at home as a child, we find
statistical associations between domestic violence and some of the other covariates included in
the regressions (not shown). We find that women who got married at younger ages are more
likely to be a victim of domestic violence. We also find that women who work, earn more
income than their partners, and are more educated that their partners are more likely to be
victims of domestic violence. This suggests that in cases where women are in a relatively
advantageous position in the relationship, their partners may rely on the use of violence as a
means to exert control over the woman and her resources. Heath (2014) has found also a positive
correlation between work and domestic violence for women with low education or young at age,
suggesting that husbands use violence to counteract working women increased bargaining power.
Interestingly, in our data we also find a positive association between women working and
earning more than their partners and the probability of women committing physical violence
against their partners.
5. Potential mechanisms
The psychology literature has found that women who grew up in violent homes are more
likely to be involved in domestic violence (Stith et al., 2000). Our evidence indicates that women
exposed to civil conflict violence are also more likely to become victims of domestic violence.
However, the mechanisms are not clear. One potential mechanism can be that women more
exposed to the conflict might regard the use of violence a normal way to resolve problems. Our
finding of an effect of exposure on the probability of physical violence from women against
their partners can be an indication on that. Following the same logic, if violence is a learned
21
social norm, women that grew up in a more violent environment may become more tolerant
towards violence from men. The DHS asks women under what circumstances they believe it is
justified for a husband to beat his wife. They were asked this hypothetical question (i.e. not about
their own husbands) before the domestic violence interview module had started. Possible reasons
for justifying beating were: “she goes out without telling the husband”; “she neglects the
children”; “she argues with husband”; “She refuses to have sex with husband;, and “she burns
the food”. In total, 4.8% of our sample said that beating was justified for any of these reasons.
We construct a variable that takes the value of 1 if the woman answered that beating is justified
for any of the previous reasons (and 0 otherwise) and estimated linear probability models similar
to equations (1), (2) and (3), but using this variable as the dependent variable. Column 1 in Table
5 shows the estimation results. We found a positive effect of early exposure to conflict violence
on the probability that women report that it is justified (for at least one reason) for husbands to
beat their wives. We also found that this effect is larger at higher levels of exposure (i.e. in the
top deciles of exposure). For instance, women in the highest decile category are 2 percentage
points more likely to justify the use of violence by men, which is a substantive effect. As a
benchmark, women who saw their fathers battered their mothers are 1.2 percentage points more
likely to justify violence against women (not shown). However, these effects are imprecisely
estimated and not statistically significant. But when we allow the effects to vary depending on
the age at the moment of exposure (Panel C, equation 3 in Table 5), we find a statistically
significant positive effect of the exposure between the ages of 9 and 16. In other words, we find
statistical evidence that women more exposed to conflict violence in their late childhood and
early teenage years are more likely to report that violence against women can be justified.
22
Further evidence in favor of the hypothesis that the use of violence can be regarded as
normal is shown in column 2 in Table 5. We estimate linear probability models similar to
equations (1), (2) and (3), but using as dependent variable the probability that a woman that
reported having been victim of domestic violence (ever) is separated from her husband (or
partner). We found that early exposure to conflict violence diminishes the probability of leaving
a violent relationship. In other words, exposure to the conflict violence increases the probability
that women who reported being a victim (ever) of domestic violence remain married (or in
cohabitation) to their partners. The effect for the average women exposed to the conflict violence
is small but, as with the other outcomes, the effects are large for women with high levels of
exposure. For example, the linear effect from equation (1) (panel A, second column of Table 5)
indicates that women exposed to 100 violent events (which corresponds to the top decile in the
distribution of exposure) are 1.8 percentage points less likely to separate from a violent
relationship. The estimates by the deciles categories of exposure in equation (2) (panel B, second
column of Table 5) are noisy and not statistically significant. Nevertheless, only exposure levels
in the top two highest deciles have a negative effect on the probability of separating from a
violent relationship, and their mean value is -1.2 percentage points. As a comparison, women
whose fathers used to beat their mothers are 1.4 percentage points less likely to leave a violent
relationship (not shown). Thus, exposure to high levels of conflict violence has an effect as large
as observing domestic violence at home while growing up. We also find larger effects of
exposure in early childhood that in teenage years on the probability of remaining in a violent
relationship (Panel C, second column of Table 5).
23
Table 5: Effects of Early Exposure to Conflict Violent Events on Justification of Violence Against
Women, on Separating from a Violent Relationship, on Marriage, and on Local Gender Ratios.
A. Equation (1)
Total CVE between 0 and 16 years old
(in hundreds)
B. Equation (2)
3rd decile
4th decile
5th decile
6th decile
7th decile
8th decile
9th decile
Highest decile
Beating of
women can be
justified
(0=No; 1=Yes)
Separated (if ever
married and
victim of
domestic abuse)
(0=No; 1=Yes)
Ever
Married
(0=No;
1=Yes)
Male/
Female
Ratio
0.008
(0.006)
-0.018**
(0.008)
-0.028
(0.022)
0.012*
(0.006)
0.006
(0.008)
0.012
(0.009)
0.017
(0.012)
0.007
(0.014)
0.010
(0.016)
0.020
(0.013)
0.020
(0.016)
0.020
(0.015)
0.001
(0.013)
0.006
(0.014)
0.018
(0.016)
0.008
(0.011)
0.004
(0.015)
0.016
(0.023)
-0.017
(0.023)
-0.006
(0.021)
-0.022
(0.020)
0.015
(0.016)
-0.007
(0.013)
-0.009
(0.016)
0.006
(0.016)
-0.020
(0.021)
-0.073**
(0.033)
-0.093***
(0.028)
0.007
(0.010)
0.003
(0.011)
0.020**
(0.009)
0.008
(0.010)
0.013
(0.010)
0.013
(0.019)
0.025
(0.019)
0.043**
(0.017)
C. Equation (3)
Number of events (in hundreds) between
0 and 8 years old
0.004
-0.027**
-0.054*
0.023**
(0.006)
(0.011)
(0.031)
(0.010)
Number of events (in hundreds) between
0.015**
-0.004
-0.037
0.021**
9 and 16 years old
(0.007)
(0.010)
(0.032)
(0.010)
Interaction term
-0.007
0.007
0.033**
-0.007*
(0.008)
(0.010)
(0.016)
(0.004)
-0.017
0.022
0.004
0.020**
Placebo: Number of events (in hundreds)
between minus 8 and 0 years old
(0.011)
(0.034)
(0.023)
(0.008)
# Observations
16,395
8,051
23,485
23,485
Notes: Standard errors are clustered at the province level. Includes only non-migrant women from
provinces affected by the internal conflict, born in 1965-1990. Each regression includes controls for
province effects, year of birth effects, and interview year effects. See notes in Table 2 for more details on
other controls included. *** denotes statistical significance at 1% level; ** denotes statistical significance
at 5% level; * denotes statistical significance at 10% level.
24
Column 3 in Table 5 shows an additional interesting finding. Women that were more
exposed to the conflict violence are less likely to get married (or cohabitate).5 Table 5 indicates
that especially women in the higher decile categories for exposure (panel B, third column) show
an important reduction in their probability of marriage. A potential explanation is that, as in the
case of the Rwandan genocide (La-Mattina, 2015) the conflict might have affected the local
marriage markets by altering gender ratios, which could have a detrimental effect in the intrahousehold bargaining power of women and therefore in the incidence of domestic violence. In
the last column of Table 5 we fitted models similar to equations (1), (2) and (3) but placing in the
left-hand side the male-to-female ratio in the same age category and province.6 We observe that
there is a positive association between higher exposure to conflict violence and the local male-tofemale ratio, contrary to the findings in the Rwandan genocide case. We posit that this finding
can be interpreted as conflict-related events being more likely to have occurred in areas with a
relative higher share of males relative to females rather than the conflict affecting the local
marriage markets. We argue this because in the case of Peru the conflict lasted for over a decade
rather than a short period of time, and the total related deaths and disappearances were not nearly
as large as in the Rwandan genocide. Moreover, our sample is comprised of women that were 15
years old or younger when the conflict started, and the fatal casualties were mostly adults. Thus,
it is unlikely that the conflict had a long-run effect on the marriage market.
Nevertheless, to test whether the impact of exposure to conflict violence runs through the
marriage market, we re-estimated a linear probability model as in equation (1) for our main
5
For these estimations we included all non-migrant women born between 1965 and 1990, not only those who are
married (or cohabitate) or were married (cohabitated) in the past.
6
Age categories were defined by intervals of 5 years, starting at 15-19 years old (then 20-24 years, etc.). The data
comes from the 2007 Peru National Census.
25
outcomes (physical violence from women to partners, any type of violence against women from
partners, probability of women justifying violence against women, separation from a violent
relationship, and marriage) but controlling for the gender ratio in the same age category and
province. The results are shown in Table 6. We find similar effects of early exposure to conflict
violence after including the local gender ratio as an additional control. In fact, with the exception
of the marriage probability, the local gender ratios are not statistically significant predictors of
the outcomes. Thus, two conclusions can be derived. First, the long-term effects of early
exposure to conflict violence on domestic violence are not explained by changes in local gender
ratios. Second, if anything, it appears that women more affected by the conflict never got
married. Thus, they are less likely to be part of our analysis sample for domestic violence (i.e.
violence from and towards the spouse or partner).
Another potential mechanism worth mentioning is that women who were more exposed
to the conflict may marry men who were also more exposed (and probably more likely see the
use of violence as normal). Unfortunately, the Peruvian DHS do not contain a male questionnaire
and therefore there is not enough information about husbands or partners to test this hypothesis.
26
Table 6: Robustness of Estimates to Controlling for Local Gender Ratio in Same Age Group
Ever
Last 12
months
Ever
Last 12
months
Total CVE between 0 and
16 years old (in hundreds)
0.010*
(0.006)
0.010**
(0.004)
0.027***
(0.008)
0.022*
(0.012)
0.008
(0.006)
Separated
(if ever
married
and
victim of
domestic
abuse)
(0=No;
1=Yes)
-0.018**
(0.008)
Male/ Female ration in age
group
0.055
(0.048)
0.027
(0.038)
0.000
(0.125)
-0.007
(0.084)
0.012
(0.039)
0.014
(0.068)
Physical violence
from woman to
partner
(0=No; 1=Yes)
Any type of violence
from partner to
woman
(0=No; 1=Yes)
Beating
of
women
can be
justified
(0=No;
1=Yes)
Ever
Married
(0=No;
1=Yes)
-0.021
(0.020)
0.637***
(0.152)
Observations
17,270
17,261
17,270
17,270
16,395
8,051
23,485
Notes: Standard errors are clustered at the province level. Includes only non-migrant women from provinces
affected by the internal conflict, born in 1965-1990. Each regression includes controls for province effects, year
of birth effects, and interview year effects. See notes in Table 2 for more details on other controls included. ***
denotes statistical significance at 1% level; ** denotes statistical significance at 5% level; * denotes statistical
significance at 10% level.
6. Robustness to model specification
Table 7 shows that after including the basic set of fixed effects to control for province,
birth cohort and interview year effects, the estimated impacts of exposure to conflict violent
events are robust to the inclusion of the other observed characteristics, including education, labor
earnings, age and age at marriage, among others. The impacts are also robust to controlling for
violence at home while growing up, which are important determinants of domestic violence.
Thus, to the extent that our findings are similar in models without with and without controls, we
are less concerned that potentially omitted confounders are significantly biasing the results.
27
Table 7: Robustness of Results to Several Model Specifications
Model Specification
Outcomes
Physical violence
from woman to
partner
(0=No; 1=Yes)
Physical violence
from partner to
woman
(0=No; 1=Yes)
Sexual violence
from partner to
woman
(0=No; 1=Yes)
Psychological
violence from
partner to woman
(0=No; 1=Yes)
Any type of
violence from
partner to woman
(0=No; 1=Yes)
I
II
III
IV
# of
observations
V
Ever
-0.004
(0.007)
0.010*
(0.006)
0.009*
(0.005)
0.010*
(0.006)
0.010*
(0.006)
17,270
Last 12
months
0.003
0.010**
0.010**
0.010**
0.010**
17,261
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
Ever
0.005
(0.011)
0.022**
(0.010)
0.018*
(0.010)
0.018*
(0.010)
0.017*
(0.010)
Last 12
months
0.003
(0.007)
0.014**
(0.006)
0.013**
(0.006)
0.012**
(0.006)
0.013**
(0.006)
Ever
0.004
(0.004)
0.003
(0.008)
0.002
(0.008)
0.003
(0.009)
0.004
(0.009)
17,269
Last 12
months
0.001
(0.003)
0.001
(0.004)
0.001
(0.004)
0.000
(0.004)
0.001
(0.004)
17,269
Ever
0.020**
(0.008)
0.020
(0.012)
0.018
(0.011)
0.019*
(0.010)
0.019*
(0.010)
17,270
Last 12
months
0.012*
(0.006)
0.021*
(0.011)
0.019*
(0.011)
0.019*
(0.011)
0.019*
(0.011)
17,267
Ever
0.013
(0.009)
0.031***
(0.012)
0.027**
(0.011)
0.027***
(0.009)
0.027***
(0.008)
17,270
Last 12
months
0.009
(0.008)
0.025**
(0.013)
0.023*
(0.012)
0.022*
(0.012)
0.022*
(0.012)
17,270
No
Yes
Yes
Yes
Yes
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
No
No
No
No
Yes
Controls
FE for birth cohort, province and
interview year
Exposure to Violence at Home
as a child
Other women and partner
characteristics
Local gender ratio
in same age group
17,270
Notes: Standard errors are clustered at the province level. Includes only non-migrant women from provinces affected
by the internal conflict, born in 1965-1990. See notes in Table 2 for more details on the list of controls available. ***
denotes statistical significance at 1% level; ** denotes statistical significance at 5% level; * denotes statistical
significance at 10% level.
28
7. Data Limitations
There are two data limitations that are worth discussing. The first one is regarding the
sample selection criteria discussed in Section 3. In particular, women in the sample must have
lived in the same town or city all their life because we cannot identify in the DHS the place of
residence during childhood and teenage years of women that migrated from somewhere else.
Information on the place of residence is crucial for matching the person with the appropriate data
on exposure to conflict violent events. Thus, for our estimation sample, we selected only those
women who reported living all their lives in the towns where they reside (i.e. non-migrant
women). By doing this, we reduced our sample size by half. This selection can introduce biases
in our estimates, and it is not clear a priori what direction the bias would take. In order to
investigate this issue further, Table 8 provides a comparison of means between migrant and nonmigrant women. Non-migrant women have on average a lower probability of being a victim of
any type of violence from their partners (although they report on average a higher incidence of
committing physical violence towards their partners). Non-migrant women have also been less
exposed on average to violence at home while growing up. Non-migrant women also got married
on average at an older age and are less likely to earn more than their partners (although they are
more likely than migrant women to be more educated than their partners). Thus, the majority of
characteristics seem to indicate that non-migrant women are less at risk of being victims of
domestic violence than migrant women. On the other hand, although it cannot be tested, it is
plausible that exposure to civil conflicts is likely to be positively related with migration (for
example, families that were displaced by the conflict). Thus, because migration is likely to be
positively correlated with the incidence of domestic violence and with the intensity of the
conflict, using only the sample of non-migrant women might introduce a downward bias in our
29
estimates.7 However, there is reason to believe that the potential bias is not severe. According to
information in the DHS, the most important reasons for migrating (among migrant women) were
family (48%), employment (27%), and education (11%). Being displaced by insecurity or
violence was only mentioned in 0.5% of the cases. Taken at face value, this suggests that at least
our selection process is not correlated with the treatment of interest (i.e. exposure to the civil
conflict), and thus our results are not likely to suffer much from this source of selection bias.
7
To see this, let’s work with the following sample selection model, where ‫ ܯ‬is the propensity to migrate, ‫ ܧܸܥ‬is
our measure of exposure to conflict violence, and (for facilitating the argument) ‫ ܸܦ‬is a continuous measure of
domestic violence. Let’s assume that ߳ଵ and ߳ଶ are jointly normally distributed with mean zero. From evidence
provided in Table 8, the covariance between ߳ଵ and ߳ଶ is likely positive (ߪଵଶ ൐ Ͳ). Also, let’s assume that ߠ ൐ Ͳ
(conflict-induced displacement effect).
‫ ܯ‬ൌ ߠ‫ ܧܸܥ‬൅ ߳ଵ
‫ ܸܦ‬ൌ ߚ‫ ܧܸܥ‬൅ ߳ଶ
We also assume that a woman migrates if ‫ ܯ‬൐ Ͳ, or equivalently if ߳ଵ ൐ െߠ‫ܧܸܥ‬. Thus, the estimation sample of
non-migrant women would consist of women for whom ߳ଵ ൑ െߠ‫ܧܸܥ‬. It can be shown that
ఙ థሺିఏ஼௏ாȀఙభ ሻ
, where ߶ሺǤ ሻ and ĭሺǤ ሻ denotes the probability density function
‫ܧ‬ሾ‫ܸܦ‬ȁ‫ܧܸܥ‬ǡ ߳ଵ ൑ െߠ‫ܧܸܥ‬ሿ ൌ ߚ‫ ܧܸܥ‬െ భమ
మ
ఙభ ĭሺିఏ஼௏ாȀఙభ ሻ
and the cumulative density function of the standard normal distribution, respectively. Since
థሺିఏ஼௏ாȀఙభ ሻ
ĭሺିఏ஼௏ாȀఙభ ሻ
is positively
correlated with ‫ܧܸܥ‬, ignoring this correction term in the regression would result in a downward bias in the estimate
of ߚ.
30
Table 8: Comparison of Means by Migration Status
Migrants
Non Migrants
P-value
of T-test
0.077
0.493
0.084
0.451
0.000
0.000
0.030
0.232
0.035
0.203
0.000
0.000
Age
Spanish speaker
No education
Incomplete primary
Complete primary
Incomplete secondary
Complete secondary
Higher Education
Worked last 12 months
No earnings from work
Woman earns more than husband
Woman earns less than husband
Woman earns about the same as husband
Other relative income category
Missing information on woman's income
Woman less educated than him
Woman more educated than him
Relative education missing
Years of marriage
Age at marriage
34.017
0.903
0.045
0.248
0.128
0.176
0.208
0.195
0.769
0.487
0.050
0.749
0.075
0.004
0.122
0.361
0.235
0.000
13.589
20.000
33.634
0.823
0.045
0.214
0.114
0.154
0.229
0.245
0.783
0.502
0.047
0.741
0.070
0.003
0.139
0.301
0.264
0.000
13.043
20.162
0.000
0.000
0.967
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.027
0.026
0.019
0.003
0.000
0.000
0.000
0.182
0.000
0.000
Exposure to Violence at Home as Child
Battered by mother
Battered by father
Father used to batter mother
0.040
0.049
0.501
0.040
0.046
0.452
0.988
0.109
0.000
Incidence of domestic violence - Ever
Physical violence from woman to partner (0=No; 1=Yes)
Any type of violence from partner to woman (0=No; 1=Yes)
Incidence of domestic violence - Last 12 months
Physical violence from woman to partner (0=No; 1=Yes)
Any type of violence from partner to woman (0=No; 1=Yes)
Women socio-demographic characteristics
Source: DHS (2004-2012)
31
The second data limitation worth discussing is measurement error in the exposure to
conflict violent events. Measuring exposure is a complicated task. We define exposure as the
number of recorded events that occurred at the woman’s province of residence. However, not all
women were affected equally by all events that occurred in their province of residence.
Unfortunately, we cannot capture that heterogeneity. Moreover, the recorded information on
conflict related events may be incomplete in a systematic way, since it relies on testimonies
gathered by the TRC. Leon (2012) argues that there may be underreporting coming from the
groups that was more affected by violence. Leon also points that since testimonies were collected
in relatively bigger cities, it may systematically undercount conflict-related events for vulnerable
populations who live in distant rural areas. This potential selective underreporting is likely to
bias our estimates, although is unclear in what direction.
8. Conclusions
We find that exposure in childhood and early teenage years to conflict violence increases
a woman’s probability of being a victim of domestic violence as an adult. The sizes of the effects
are small for the average woman affected by the conflict. However, the average effects mask
important heterogeneities as some provinces were affected by the conflict more severely than
others. We find that the effects of exposure to the conflict violence are large for women in the
highest categories of exposure. We also find evidence that a potential mechanism through which
exposure to the conflict affects domestic violence in the long-term is normalization of the use of
violence. Women more exposed to conflict violent events are also more likely to be perpetrator
of violence, more likely to report that violence toward women can be justified, and more likely to
stay in a violent relationship.
32
Thus, our study indicates that civil conflicts may have long-term effects, not only on
educational attainment, health status and earnings, as it has been documented in other studies
(Galdo, 2013, Grimard and Laszlo, 2014, Leon, 2009), but also on increasing the level of
domestic violence in the society. Moreover, our findings and previous research in psychology
(Barnett, Miller-Perrin and Perrin, 2005, Caesar, 1988, Cappell and Heiner, 1990, Mihalic and
Elliott, 1997, Simons et al., 1993, Stith et al., 2000) have found that individuals that witness
domestic violence at home when growing up are more likely to experience it in their own
marriages. This would imply that civil conflicts can increase the level of domestic violence not
only for the generation that was directly exposed to it but also for future generations as well.
Potential avenues for further research should include studying the impact of civil conflicts on
other forms of violence and aggressive behavior in society, and evaluating policy alternatives to
prevent the intergenerational transmission of the violent patterns that this paper suggests can
occur.
Acknowledgements
We thank Jeffrey Smith, H. Luke Shaefer and Anne Fitzpatrick for comments on earlier
drafts of this paper and Tanya Byker, Andrew Goodman-Bacon, Robert Garlick, Rosanna Lee,
seminar participants at the University of Michigan and participants at the 2010 Northeast
Universities Development Consortium (NEUDC) Conference for helpful suggestions. We also
thank the participants at a seminar in the University of Frankfurt am Mainz. We also thank
Gianmarco Leon for providing us with the data on the civil conflict events compiled by the Truth
and Reconciliation Commission. All errors are our own.
33
References
Barnett, Ola; Cindy L. Miller-Perrin and Robin D. Perrin. 2005. Family Violence across the
Lifespan. An Introduction. Thousand Oaks, California: Sage Publications, Inc.
Buvinic, Mayra; Andrew R. Morrison and Michael Shifter. 1999. "Violence in the Americas: A
Framework for Action," A. R. Morrison and M. L. Biehl, Too Close to Home: Domestic
Violence in the Americas. Washington, D.C.: Inter-American Development Bank,
Caesar, P. Lynn. 1988. "Exposure to Violence in the Families-of-Origin among Wife-Abusers
and Maritally Nonviolent Men." Violence and Victims, 3(1), 49-63.
Cappell, Charles and Robert Heiner. 1990. "The Intergenerational Transimission of Family
Agression." Journal of Family Violence, 5(2), 135-52.
ECLAC. 2015. "Annual Report 2013-2014. Confronting Violence against Women in Latin
America and the Caribbean (Lc/G.2626)," Santiago, Chile: Economic Commission for Latin
America and the Caribbean,
Fowler, Patrick J.; Carolyn J. Tompsett; Jordan M. Braciszewski; Angela J. Jaques-Tiura and
Boris B. Baltes. 2009. "Community Violence: A Meta-Analysis on the Effect of Exposure
and Mental Health Outcomes of Children and Adolescents." Development and
Psychopathology, 21, 227-59.
Galdo, Jose. 2010. "The Long-Run Labor Market Consequences of Civil War: Evidence from the
Shining Path in Peru." Economic Development and Cultural Change, 61(4), 789-823.
____. 2013. "The Long-Run Labor Market Consequences of Civil War: Evidence from the
Shining Path in Peru." Economic Development and Cultural Change, 61(4), 789-823.
Grimard, Franque and Sonia Laszlo. 2014. "Long-Term Effects of Civil Conflict on Women’s
Health Outcomes in Peru." World Development, 54, 139-55.
Guerra, Nancy G.; L. Rowell Huesmann and Anja Spindler. 2003. "Community Violence
Exposure, Social Cognition, and Aggression among Urban Elementary School Children."
Child Development, 74, 1561-76.
Heath, Rachel. 2014. "Women’s Access to Labor Market Opportunities, Control of Household
Resources, and Domestic Violence: Evidence from Bangladesh." World Development, 57,
32-46.
34
La-Mattina, Giulia. 2015. "Civil Conflict, Domestic Violence and Intra-Household Bargaining in
Post-Genocide Rwanda." Journal of Development Economics, Forthcoming.
Leon, Gianmarco. 2012. "Civil Conflict and Human Capital Accumulation the Long-Term
Effects of Political Violence in Perú." Journal of Human Resources, 47(4), 991-1022.
____. 2009. "Civil Conflict and Human Capital Accumulation: The Long-Term Effects of
Political Violence in Peru," BREAD Working Paper No.245.
Mihalic, Sharon Wofford and Delbert Elliott. 1997. "A Social Learning Theory Model of Marital
Violence." Journal of Family Violence, 12(1), 21-47.
Noe, Dominik and Johannes Rieckmann. 2013. "Violent Behaviour: The Effect of Civil Conflict
on Domestic Violence in Colombia." Courant Research Centre: Poverty, Equity, and
Growth--Discussion Papers, (136).
Sanchez, Alan. 2010. "The Impact of Armed Conflict on Early Health and Nutrition in Peru,"
Working Paper.
Scarpa, Angela; Sara Chiara Haden and Jimmy Hurley. 2006. "Community Violence
Victimization and Symptoms of Posttraumatic Stress Disorder: The Moderating Effects of
Coping and Social Support." Journal of Interpersonal Violence, 21, 446-69.
Scarpa, Angela; Jimmy D. Hurley; Howard W. Shumate and Sara Chiara Haden. 2006. "Lifetime
Prevalence and Socioemotional Effects of Hearing About Community Violence." Journal of
Interpersonal Violence, 21(1), 5-23.
Schwab-Stone, Mary E.; Tim S. Ayers; Wesley Kasprow; Charlene Voyce; Charles Barone;
Timothy Shriver and Roger P. Weissberg. 1995. "No Safe Haven: A Study of Violence
Exposure in an Urban Community." Journal of the American Academy of Child and
Adolescent Psychiatry, 34, 1343-52.
Simons, Ronald R.; Christine Johnson; Jay Beaman and Rand D. Conger. 1993. "Explaining
Women's Double Jeopardy: Factors That Mediate the Association between Harsh Treatment
as a Child and Violence by a Husband." Journal of Marriage and Family, 55(3), 713-23.
Stith, Sandra; Karen H. Rosen; Kimberly A. Middleton; Amy L. Busch; Kirsten Lundeberg and
Russell P. Carlton. 2000. "The Intergenerational Transmission of Spouse Abuse: A MetaAnalysis." Journal of Marriage and Family, 62(3), 640-54.
35
TRC.
2004.
"Final
Report,"
Truth
and
Reconciliation
Commission
-
Peru,
http://www.cverdad.org.pe/ingles/ifinal/index.php.
Williams, Richard. 2007. "The Psychosocial Consequences for Children of Mass Violence,
Terrorism and Disasters." International Review of Psychiatry, 19(3), 263-77.
36
Appendix: Domestic Violence Questions in the Peru Demographic and Health Surveys
A. Physical violence from woman to partner.
x
Have you ever hit, slapped, kicked, or done anything else to physically hurt your (last)
(husband/partner) at times when he was not already beating or physically hurting you?
(0=no; 1=yes)
x
In the last 12 months, how often have you done this to your (last) (husband/partner):
(0=not at all; 1=often or only sometimes)
B. Psychological violence from partner to woman
If the woman responded “Yes” to any of the questions below we code her as having been
victim of psychological violence at some point (ever).
If she responded “Often” or “Only sometimes” to any of the questions below regarding the
last 12 months, we code her as having been victim of psychological violence in the last 12
months.
Did your (last) (husband/partner):
Ever
How often did this happen
during the last 12 months:
Say or do something to humiliate you in front of
others?
Yes/No
Often/ only sometimes/ not at all?
Threaten to hurt or harm you or someone you care
about?
Yes/No
Often/ only sometimes/r not at
all?
Insult you or make you feel bad about yourself?
Yes/No
Often/ only sometimes/r not at
all?
C. Physical violence from partner to woman
If the woman responded “Yes” to any of the questions below we code her as having been
victim of physical violence at some point (ever).
If she responded “Often” or “Only sometimes” to any of the questions below regarding the
last 12 months, we code her as having been victim of physical violence in the last 12 months.
37
Did your (last) (husband/partner) do any of the
following things to you:
Ever
How often did this happen
during the last 12 months:
Push you, shake you, or throw something at you?
Yes/No
Often/ only sometimes/not at all?
Slap you or twist your arm?
Yes/No
Often/ only sometimes/not at all?
Punch you with his fist or with something that
could hurt you?
Yes/No
Often/ only sometimes/not at all?
Kick you, drag you, or beat you up?
Yes/No
Often/ only sometimes/not at all?
Try to choke you or burn you on purpose?
Yes/No
Often/ only sometimes/not at all?
Attack you with a knife, gun, or other weapon?
Yes/No
Often/ only sometimes/not at all?
Threaten you with a knife, gun, or other weapon?
Yes/No
Often/ only sometimes/not at all?
D. Sexual violence from partner to woman
If the woman responded “Yes” to any of the questions below we code her as having been
victim of sexual violence at some point (ever).
If she responded “Often” or “Only sometimes” to any of the questions below regarding the
last 12 months, we code her as having been victim of sexual violence in the last 12 months.
Did your (last) (husband/partner) do any of the
following things to you:
Ever
How often did this happen
during the last 12 months:
Physically force you to have sexual intercourse with Yes/No
him when you did not want to?
Often/ only sometimes/not at all?
Force you to perform any other sexual acts you did
not want to?
Often/ only sometimes/not at all?
38
Yes/No